Integrating Semi-supervised and Supervised Learning Methods for Label Fusion in Multi-Atlas Based Image Segmentation.

Area of Science:

• Medical Image Analysis
• Machine Learning
• Pattern Recognition

Background:

• Multi-atlas based image segmentation is crucial for medical image analysis.
• Existing methods often require extensive manual annotation or struggle with segmentation accuracy.
• Integrating diverse machine learning techniques can enhance segmentation robustness.

Purpose of the Study:

• To develop a novel label fusion method for multi-atlas image segmentation.
• To improve segmentation accuracy by combining semi-supervised and supervised learning within a pattern recognition framework.
• To evaluate the proposed method's performance in segmenting hippocampus in MR images.

Main Methods:

• A pattern recognition-based multi-atlas label fusion framework was developed.
• Random forests classification models were built using registered atlas image patches for each voxel.
• A semi-supervised label propagation method refined the probabilistic segmentation map, considering local and global image appearance consistency.

Main Results:

• The proposed method achieved competitive segmentation performance for hippocampus MR images.
• The average Dice index exceeded 0.88, demonstrating effective segmentation.
• Performance was comparable to alternative machine learning-based multi-atlas segmentation methods.

Conclusions:

• The novel label fusion method effectively integrates semi-supervised and supervised learning for enhanced multi-atlas image segmentation.
• The approach demonstrates robust performance in segmenting anatomical structures like the hippocampus.
• Publicly available source codes facilitate reproducibility and further research.